Messages in this thread | | | Date | Wed, 10 Sep 2008 14:28:21 -0400 | From | Oren Laadan <> | Subject | Re: [RFC v4][PATCH 4/9] Memory management (dump) |
| |
Dave Hansen wrote: > On Tue, 2008-09-09 at 03:42 -0400, Oren Laadan wrote: >> + while (addr < end) { >> + struct page *page; >> + >> + /* >> + * simplified version of get_user_pages(): already have vma, >> + * only need FOLL_TOUCH, and (for now) ignore fault stats. >> + * >> + * FIXME: consolidate with get_user_pages() >> + */ >> + >> + cond_resched(); >> + while (!(page = follow_page(vma, addr, FOLL_TOUCH))) { >> + ret = handle_mm_fault(vma->vm_mm, vma, addr, 0); >> + if (ret & VM_FAULT_ERROR) { >> + if (ret & VM_FAULT_OOM) >> + ret = -ENOMEM; >> + else if (ret & VM_FAULT_SIGBUS) >> + ret = -EFAULT; >> + else >> + BUG(); >> + break; >> + } >> + cond_resched(); >> + ret = 0; >> + } > > get_user_pages() is really the wrong thing to use here. It makes pages > *present* so that we can do things like hand them off to a driver. For > checkpointing, we really don't care about that. It's a waste of time, > for instance to perform faults to fill the mappings up with zero pages > and page tables. Just think of what will happen the first time we touch > a very large, very sparse anonymous area. We'll probably kill the > system just allocating page tables. Take a look at the comment in > follow_page(). This is a similar operation to core dumping, and we need > to be careful. > > This might be fine for a proof of concept, but it needs to be thought > out much more thoroughly before getting merged. I guess I'm > volunteering to go do that.
The intention is not to allocate unallocated pages, but to get the page pointer and bring in swapped out pages if necessary. (Avoiding swap-in is possible, but left for future optimization).
Indeed, follow_page() does the work just fine; Of course, it should be called with FOLL_ANON instead of FOLL_TOUCH. Thanks for pointing out.
Oren.
| |